drivers/net/hyperv/netvsc_drv.c

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, see <http://www.gnu.org/licenses/>.
  15  *
  16  * Authors:
  17  *   Haiyang Zhang <haiyangz@microsoft.com>
  18  *   Hank Janssen  <hjanssen@microsoft.com>
  19  */
  20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  21
  22 #include <linux/init.h>
  23 #include <linux/atomic.h>
  24 #include <linux/module.h>
  25 #include <linux/highmem.h>
  26 #include <linux/device.h>
  27 #include <linux/io.h>
  28 #include <linux/delay.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/inetdevice.h>
  31 #include <linux/etherdevice.h>
  32 #include <linux/skbuff.h>
  33 #include <linux/if_vlan.h>
  34 #include <linux/in.h>
  35 #include <linux/slab.h>
  36 #include <linux/rtnetlink.h>
  37 #include <linux/netpoll.h>
  38 #include <linux/reciprocal_div.h>
  39
  40 #include <net/arp.h>
  41 #include <net/route.h>
  42 #include <net/sock.h>
  43 #include <net/pkt_sched.h>
  44 #include <net/checksum.h>
  45 #include <net/ip6_checksum.h>
  46
  47 #include "hyperv_net.h"
  48
  49 #define RING_SIZE_MIN           64
  50
  51 #define LINKCHANGE_INT (2 * HZ)
  52 #define VF_TAKEOVER_INT (HZ / 10)
  53
  54 static unsigned int ring_size __ro_after_init = 128;
  55 module_param(ring_size, uint, S_IRUGO);
  56 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  57 unsigned int netvsc_ring_bytes __ro_after_init;
  58 struct reciprocal_value netvsc_ring_reciprocal __ro_after_init;
  59
  60 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
  61                                 NETIF_MSG_LINK | NETIF_MSG_IFUP |
  62                                 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
  63                                 NETIF_MSG_TX_ERR;
  64
  65 static int debug = -1;
  66 module_param(debug, int, S_IRUGO);
  67 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  68
  69 static void netvsc_set_multicast_list(struct net_device *net)
  70 {
  71         struct net_device_context *net_device_ctx = netdev_priv(net);
  72         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
  73
  74         rndis_filter_update(nvdev);
  75 }
  76
  77 static int netvsc_open(struct net_device *net)
  78 {
  79         struct net_device_context *ndev_ctx = netdev_priv(net);
  80         struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
  81         struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
  82         struct rndis_device *rdev;
  83         int ret = 0;
  84
  85         netif_carrier_off(net);
  86
  87         /* Open up the device */
  88         ret = rndis_filter_open(nvdev);
  89         if (ret != 0) {
  90                 netdev_err(net, "unable to open device (ret %d).\n", ret);
  91                 return ret;
  92         }
  93
  94         netif_tx_wake_all_queues(net);
  95
  96         rdev = nvdev->extension;
  97
  98         if (!rdev->link_state)
  99                 netif_carrier_on(net);
 100
 101         if (vf_netdev) {
 102                 /* Setting synthetic device up transparently sets
 103                  * slave as up. If open fails, then slave will be
 104                  * still be offline (and not used).
 105                  */
 106                 ret = dev_open(vf_netdev);
 107                 if (ret)
 108                         netdev_warn(net,
 109                                     "unable to open slave: %s: %d\n",
 110                                     vf_netdev->name, ret);
 111         }
 112         return 0;
 113 }
 114
 115 static int netvsc_close(struct net_device *net)
 116 {
 117         struct net_device_context *net_device_ctx = netdev_priv(net);
 118         struct net_device *vf_netdev
 119                 = rtnl_dereference(net_device_ctx->vf_netdev);
 120         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 121         int ret = 0;
 122         u32 aread, i, msec = 10, retry = 0, retry_max = 20;
 123         struct vmbus_channel *chn;
 124
 125         netif_tx_disable(net);
 126
 127         /* No need to close rndis filter if it is removed already */
 128         if (!nvdev)
 129                 goto out;
 130
 131         ret = rndis_filter_close(nvdev);
 132         if (ret != 0) {
 133                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 134                 return ret;
 135         }
 136
 137         /* Ensure pending bytes in ring are read */
 138         while (true) {
 139                 aread = 0;
 140                 for (i = 0; i < nvdev->num_chn; i++) {
 141                         chn = nvdev->chan_table[i].channel;
 142                         if (!chn)
 143                                 continue;
 144
 145                         aread = hv_get_bytes_to_read(&chn->inbound);
 146                         if (aread)
 147                                 break;
 148
 149                         aread = hv_get_bytes_to_read(&chn->outbound);
 150                         if (aread)
 151                                 break;
 152                 }
 153
 154                 retry++;
 155                 if (retry > retry_max || aread == 0)
 156                         break;
 157
 158                 msleep(msec);
 159
 160                 if (msec < 1000)
 161                         msec *= 2;
 162         }
 163
 164         if (aread) {
 165                 netdev_err(net, "Ring buffer not empty after closing rndis\n");
 166                 ret = -ETIMEDOUT;
 167         }
 168
 169 out:
 170         if (vf_netdev)
 171                 dev_close(vf_netdev);
 172
 173         return ret;
 174 }
 175
 176 static inline void *init_ppi_data(struct rndis_message *msg,
 177                                   u32 ppi_size, u32 pkt_type)
 178 {
 179         struct rndis_packet *rndis_pkt = &msg->msg.pkt;
 180         struct rndis_per_packet_info *ppi;
 181
 182         rndis_pkt->data_offset += ppi_size;
 183         ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset
 184                 + rndis_pkt->per_pkt_info_len;
 185
 186         ppi->size = ppi_size;
 187         ppi->type = pkt_type;
 188         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
 189
 190         rndis_pkt->per_pkt_info_len += ppi_size;
 191
 192         return ppi + 1;
 193 }
 194
 195 /* Azure hosts don't support non-TCP port numbers in hashing for fragmented
 196  * packets. We can use ethtool to change UDP hash level when necessary.
 197  */
 198 static inline u32 netvsc_get_hash(
 199         struct sk_buff *skb,
 200         const struct net_device_context *ndc)
 201 {
 202         struct flow_keys flow;
 203         u32 hash, pkt_proto = 0;
 204         static u32 hashrnd __read_mostly;
 205
 206         net_get_random_once(&hashrnd, sizeof(hashrnd));
 207
 208         if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
 209                 return 0;
 210
 211         switch (flow.basic.ip_proto) {
 212         case IPPROTO_TCP:
 213                 if (flow.basic.n_proto == htons(ETH_P_IP))
 214                         pkt_proto = HV_TCP4_L4HASH;
 215                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 216                         pkt_proto = HV_TCP6_L4HASH;
 217
 218                 break;
 219
 220         case IPPROTO_UDP:
 221                 if (flow.basic.n_proto == htons(ETH_P_IP))
 222                         pkt_proto = HV_UDP4_L4HASH;
 223                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 224                         pkt_proto = HV_UDP6_L4HASH;
 225
 226                 break;
 227         }
 228
 229         if (pkt_proto & ndc->l4_hash) {
 230                 return skb_get_hash(skb);
 231         } else {
 232                 if (flow.basic.n_proto == htons(ETH_P_IP))
 233                         hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
 234                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 235                         hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
 236                 else
 237                         hash = 0;
 238
 239                 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
 240         }
 241
 242         return hash;
 243 }
 244
 245 static inline int netvsc_get_tx_queue(struct net_device *ndev,
 246                                       struct sk_buff *skb, int old_idx)
 247 {
 248         const struct net_device_context *ndc = netdev_priv(ndev);
 249         struct sock *sk = skb->sk;
 250         int q_idx;
 251
 252         q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
 253                               (VRSS_SEND_TAB_SIZE - 1)];
 254
 255         /* If queue index changed record the new value */
 256         if (q_idx != old_idx &&
 257             sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
 258                 sk_tx_queue_set(sk, q_idx);
 259
 260         return q_idx;
 261 }
 262
 263 /*
 264  * Select queue for transmit.
 265  *
 266  * If a valid queue has already been assigned, then use that.
 267  * Otherwise compute tx queue based on hash and the send table.
 268  *
 269  * This is basically similar to default (__netdev_pick_tx) with the added step
 270  * of using the host send_table when no other queue has been assigned.
 271  *
 272  * TODO support XPS - but get_xps_queue not exported
 273  */
 274 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
 275 {
 276         int q_idx = sk_tx_queue_get(skb->sk);
 277
 278         if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
 279                 /* If forwarding a packet, we use the recorded queue when
 280                  * available for better cache locality.
 281                  */
 282                 if (skb_rx_queue_recorded(skb))
 283                         q_idx = skb_get_rx_queue(skb);
 284                 else
 285                         q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
 286         }
 287
 288         return q_idx;
 289 }
 290
 291 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 292                                void *accel_priv,
 293                                select_queue_fallback_t fallback)
 294 {
 295         struct net_device_context *ndc = netdev_priv(ndev);
 296         struct net_device *vf_netdev;
 297         u16 txq;
 298
 299         rcu_read_lock();
 300         vf_netdev = rcu_dereference(ndc->vf_netdev);
 301         if (vf_netdev) {
 302                 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
 303                 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
 304         } else {
 305                 txq = netvsc_pick_tx(ndev, skb);
 306         }
 307         rcu_read_unlock();
 308
 309         while (unlikely(txq >= ndev->real_num_tx_queues))
 310                 txq -= ndev->real_num_tx_queues;
 311
 312         return txq;
 313 }
 314
 315 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
 316                        struct hv_page_buffer *pb)
 317 {
 318         int j = 0;
 319
 320         /* Deal with compund pages by ignoring unused part
 321          * of the page.
 322          */
 323         page += (offset >> PAGE_SHIFT);
 324         offset &= ~PAGE_MASK;
 325
 326         while (len > 0) {
 327                 unsigned long bytes;
 328
 329                 bytes = PAGE_SIZE - offset;
 330                 if (bytes > len)
 331                         bytes = len;
 332                 pb[j].pfn = page_to_pfn(page);
 333                 pb[j].offset = offset;
 334                 pb[j].len = bytes;
 335
 336                 offset += bytes;
 337                 len -= bytes;
 338
 339                 if (offset == PAGE_SIZE && len) {
 340                         page++;
 341                         offset = 0;
 342                         j++;
 343                 }
 344         }
 345
 346         return j + 1;
 347 }
 348
 349 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 350                            struct hv_netvsc_packet *packet,
 351                            struct hv_page_buffer *pb)
 352 {
 353         u32 slots_used = 0;
 354         char *data = skb->data;
 355         int frags = skb_shinfo(skb)->nr_frags;
 356         int i;
 357
 358         /* The packet is laid out thus:
 359          * 1. hdr: RNDIS header and PPI
 360          * 2. skb linear data
 361          * 3. skb fragment data
 362          */
 363         slots_used += fill_pg_buf(virt_to_page(hdr),
 364                                   offset_in_page(hdr),
 365                                   len, &pb[slots_used]);
 366
 367         packet->rmsg_size = len;
 368         packet->rmsg_pgcnt = slots_used;
 369
 370         slots_used += fill_pg_buf(virt_to_page(data),
 371                                 offset_in_page(data),
 372                                 skb_headlen(skb), &pb[slots_used]);
 373
 374         for (i = 0; i < frags; i++) {
 375                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 376
 377                 slots_used += fill_pg_buf(skb_frag_page(frag),
 378                                         frag->page_offset,
 379                                         skb_frag_size(frag), &pb[slots_used]);
 380         }
 381         return slots_used;
 382 }
 383
 384 static int count_skb_frag_slots(struct sk_buff *skb)
 385 {
 386         int i, frags = skb_shinfo(skb)->nr_frags;
 387         int pages = 0;
 388
 389         for (i = 0; i < frags; i++) {
 390                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 391                 unsigned long size = skb_frag_size(frag);
 392                 unsigned long offset = frag->page_offset;
 393
 394                 /* Skip unused frames from start of page */
 395                 offset &= ~PAGE_MASK;
 396                 pages += PFN_UP(offset + size);
 397         }
 398         return pages;
 399 }
 400
 401 static int netvsc_get_slots(struct sk_buff *skb)
 402 {
 403         char *data = skb->data;
 404         unsigned int offset = offset_in_page(data);
 405         unsigned int len = skb_headlen(skb);
 406         int slots;
 407         int frag_slots;
 408
 409         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
 410         frag_slots = count_skb_frag_slots(skb);
 411         return slots + frag_slots;
 412 }
 413
 414 static u32 net_checksum_info(struct sk_buff *skb)
 415 {
 416         if (skb->protocol == htons(ETH_P_IP)) {
 417                 struct iphdr *ip = ip_hdr(skb);
 418
 419                 if (ip->protocol == IPPROTO_TCP)
 420                         return TRANSPORT_INFO_IPV4_TCP;
 421                 else if (ip->protocol == IPPROTO_UDP)
 422                         return TRANSPORT_INFO_IPV4_UDP;
 423         } else {
 424                 struct ipv6hdr *ip6 = ipv6_hdr(skb);
 425
 426                 if (ip6->nexthdr == IPPROTO_TCP)
 427                         return TRANSPORT_INFO_IPV6_TCP;
 428                 else if (ip6->nexthdr == IPPROTO_UDP)
 429                         return TRANSPORT_INFO_IPV6_UDP;
 430         }
 431
 432         return TRANSPORT_INFO_NOT_IP;
 433 }
 434
 435 /* Send skb on the slave VF device. */
 436 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
 437                           struct sk_buff *skb)
 438 {
 439         struct net_device_context *ndev_ctx = netdev_priv(net);
 440         unsigned int len = skb->len;
 441         int rc;
 442
 443         skb->dev = vf_netdev;
 444         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 445
 446         rc = dev_queue_xmit(skb);
 447         if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
 448                 struct netvsc_vf_pcpu_stats *pcpu_stats
 449                         = this_cpu_ptr(ndev_ctx->vf_stats);
 450
 451                 u64_stats_update_begin(&pcpu_stats->syncp);
 452                 pcpu_stats->tx_packets++;
 453                 pcpu_stats->tx_bytes += len;
 454                 u64_stats_update_end(&pcpu_stats->syncp);
 455         } else {
 456                 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
 457         }
 458
 459         return rc;
 460 }
 461
 462 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 463 {
 464         struct net_device_context *net_device_ctx = netdev_priv(net);
 465         struct hv_netvsc_packet *packet = NULL;
 466         int ret;
 467         unsigned int num_data_pgs;
 468         struct rndis_message *rndis_msg;
 469         struct net_device *vf_netdev;
 470         u32 rndis_msg_size;
 471         u32 hash;
 472         struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
 473
 474         /* if VF is present and up then redirect packets
 475          * already called with rcu_read_lock_bh
 476          */
 477         vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
 478         if (vf_netdev && netif_running(vf_netdev) &&
 479             !netpoll_tx_running(net))
 480                 return netvsc_vf_xmit(net, vf_netdev, skb);
 481
 482         /* We will atmost need two pages to describe the rndis
 483          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 484          * of pages in a single packet. If skb is scattered around
 485          * more pages we try linearizing it.
 486          */
 487
 488         num_data_pgs = netvsc_get_slots(skb) + 2;
 489
 490         if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
 491                 ++net_device_ctx->eth_stats.tx_scattered;
 492
 493                 if (skb_linearize(skb))
 494                         goto no_memory;
 495
 496                 num_data_pgs = netvsc_get_slots(skb) + 2;
 497                 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 498                         ++net_device_ctx->eth_stats.tx_too_big;
 499                         goto drop;
 500                 }
 501         }
 502
 503         /*
 504          * Place the rndis header in the skb head room and
 505          * the skb->cb will be used for hv_netvsc_packet
 506          * structure.
 507          */
 508         ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
 509         if (ret)
 510                 goto no_memory;
 511
 512         /* Use the skb control buffer for building up the packet */
 513         BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
 514                         FIELD_SIZEOF(struct sk_buff, cb));
 515         packet = (struct hv_netvsc_packet *)skb->cb;
 516
 517         packet->q_idx = skb_get_queue_mapping(skb);
 518
 519         packet->total_data_buflen = skb->len;
 520         packet->total_bytes = skb->len;
 521         packet->total_packets = 1;
 522
 523         rndis_msg = (struct rndis_message *)skb->head;
 524
 525         /* Add the rndis header */
 526         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 527         rndis_msg->msg_len = packet->total_data_buflen;
 528
 529         rndis_msg->msg.pkt = (struct rndis_packet) {
 530                 .data_offset = sizeof(struct rndis_packet),
 531                 .data_len = packet->total_data_buflen,
 532                 .per_pkt_info_offset = sizeof(struct rndis_packet),
 533         };
 534
 535         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 536
 537         hash = skb_get_hash_raw(skb);
 538         if (hash != 0 && net->real_num_tx_queues > 1) {
 539                 u32 *hash_info;
 540
 541                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 542                 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 543                                           NBL_HASH_VALUE);
 544                 *hash_info = hash;
 545         }
 546
 547         if (skb_vlan_tag_present(skb)) {
 548                 struct ndis_pkt_8021q_info *vlan;
 549
 550                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 551                 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 552                                      IEEE_8021Q_INFO);
 553
 554                 vlan->value = 0;
 555                 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
 556                 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
 557                                 VLAN_PRIO_SHIFT;
 558         }
 559
 560         if (skb_is_gso(skb)) {
 561                 struct ndis_tcp_lso_info *lso_info;
 562
 563                 rndis_msg_size += NDIS_LSO_PPI_SIZE;
 564                 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 565                                          TCP_LARGESEND_PKTINFO);
 566
 567                 lso_info->value = 0;
 568                 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 569                 if (skb->protocol == htons(ETH_P_IP)) {
 570                         lso_info->lso_v2_transmit.ip_version =
 571                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 572                         ip_hdr(skb)->tot_len = 0;
 573                         ip_hdr(skb)->check = 0;
 574                         tcp_hdr(skb)->check =
 575                                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 576                                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 577                 } else {
 578                         lso_info->lso_v2_transmit.ip_version =
 579                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 580                         ipv6_hdr(skb)->payload_len = 0;
 581                         tcp_hdr(skb)->check =
 582                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 583                                                  &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 584                 }
 585                 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
 586                 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 587         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 588                 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
 589                         struct ndis_tcp_ip_checksum_info *csum_info;
 590
 591                         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 592                         csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 593                                                   TCPIP_CHKSUM_PKTINFO);
 594
 595                         csum_info->value = 0;
 596                         csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
 597
 598                         if (skb->protocol == htons(ETH_P_IP)) {
 599                                 csum_info->transmit.is_ipv4 = 1;
 600
 601                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
 602                                         csum_info->transmit.tcp_checksum = 1;
 603                                 else
 604                                         csum_info->transmit.udp_checksum = 1;
 605                         } else {
 606                                 csum_info->transmit.is_ipv6 = 1;
 607
 608                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 609                                         csum_info->transmit.tcp_checksum = 1;
 610                                 else
 611                                         csum_info->transmit.udp_checksum = 1;
 612                         }
 613                 } else {
 614                         /* Can't do offload of this type of checksum */
 615                         if (skb_checksum_help(skb))
 616                                 goto drop;
 617                 }
 618         }
 619
 620         /* Start filling in the page buffers with the rndis hdr */
 621         rndis_msg->msg_len += rndis_msg_size;
 622         packet->total_data_buflen = rndis_msg->msg_len;
 623         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 624                                                skb, packet, pb);
 625
 626         /* timestamp packet in software */
 627         skb_tx_timestamp(skb);
 628
 629         ret = netvsc_send(net, packet, rndis_msg, pb, skb);
 630         if (likely(ret == 0))
 631                 return NETDEV_TX_OK;
 632
 633         if (ret == -EAGAIN) {
 634                 ++net_device_ctx->eth_stats.tx_busy;
 635                 return NETDEV_TX_BUSY;
 636         }
 637
 638         if (ret == -ENOSPC)
 639                 ++net_device_ctx->eth_stats.tx_no_space;
 640
 641 drop:
 642         dev_kfree_skb_any(skb);
 643         net->stats.tx_dropped++;
 644
 645         return NETDEV_TX_OK;
 646
 647 no_memory:
 648         ++net_device_ctx->eth_stats.tx_no_memory;
 649         goto drop;
 650 }
 651
 652 /*
 653  * netvsc_linkstatus_callback - Link up/down notification
 654  */
 655 void netvsc_linkstatus_callback(struct net_device *net,
 656                                 struct rndis_message *resp)
 657 {
 658         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 659         struct net_device_context *ndev_ctx = netdev_priv(net);
 660         struct netvsc_reconfig *event;
 661         unsigned long flags;
 662
 663         /* Update the physical link speed when changing to another vSwitch */
 664         if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
 665                 u32 speed;
 666
 667                 speed = *(u32 *)((void *)indicate
 668                                  + indicate->status_buf_offset) / 10000;
 669                 ndev_ctx->speed = speed;
 670                 return;
 671         }
 672
 673         /* Handle these link change statuses below */
 674         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
 675             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
 676             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
 677                 return;
 678
 679         if (net->reg_state != NETREG_REGISTERED)
 680                 return;
 681
 682         event = kzalloc(sizeof(*event), GFP_ATOMIC);
 683         if (!event)
 684                 return;
 685         event->event = indicate->status;
 686
 687         spin_lock_irqsave(&ndev_ctx->lock, flags);
 688         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
 689         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
 690
 691         schedule_delayed_work(&ndev_ctx->dwork, 0);
 692 }
 693
 694 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
 695                                              struct napi_struct *napi,
 696                                              const struct ndis_tcp_ip_checksum_info *csum_info,
 697                                              const struct ndis_pkt_8021q_info *vlan,
 698                                              void *data, u32 buflen)
 699 {
 700         struct sk_buff *skb;
 701
 702         skb = napi_alloc_skb(napi, buflen);
 703         if (!skb)
 704                 return skb;
 705
 706         /*
 707          * Copy to skb. This copy is needed here since the memory pointed by
 708          * hv_netvsc_packet cannot be deallocated
 709          */
 710         skb_put_data(skb, data, buflen);
 711
 712         skb->protocol = eth_type_trans(skb, net);
 713
 714         /* skb is already created with CHECKSUM_NONE */
 715         skb_checksum_none_assert(skb);
 716
 717         /*
 718          * In Linux, the IP checksum is always checked.
 719          * Do L4 checksum offload if enabled and present.
 720          */
 721         if (csum_info && (net->features & NETIF_F_RXCSUM)) {
 722                 if (csum_info->receive.tcp_checksum_succeeded ||
 723                     csum_info->receive.udp_checksum_succeeded)
 724                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 725         }
 726
 727         if (vlan) {
 728                 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT);
 729
 730                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 731                                        vlan_tci);
 732         }
 733
 734         return skb;
 735 }
 736
 737 /*
 738  * netvsc_recv_callback -  Callback when we receive a packet from the
 739  * "wire" on the specified device.
 740  */
 741 int netvsc_recv_callback(struct net_device *net,
 742                          struct netvsc_device *net_device,
 743                          struct vmbus_channel *channel,
 744                          void  *data, u32 len,
 745                          const struct ndis_tcp_ip_checksum_info *csum_info,
 746                          const struct ndis_pkt_8021q_info *vlan)
 747 {
 748         struct net_device_context *net_device_ctx = netdev_priv(net);
 749         u16 q_idx = channel->offermsg.offer.sub_channel_index;
 750         struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
 751         struct sk_buff *skb;
 752         struct netvsc_stats *rx_stats;
 753
 754         if (net->reg_state != NETREG_REGISTERED)
 755                 return NVSP_STAT_FAIL;
 756
 757         /* Allocate a skb - TODO direct I/O to pages? */
 758         skb = netvsc_alloc_recv_skb(net, &nvchan->napi,
 759                                     csum_info, vlan, data, len);
 760         if (unlikely(!skb)) {
 761                 ++net_device_ctx->eth_stats.rx_no_memory;
 762                 rcu_read_unlock();
 763                 return NVSP_STAT_FAIL;
 764         }
 765
 766         skb_record_rx_queue(skb, q_idx);
 767
 768         /*
 769          * Even if injecting the packet, record the statistics
 770          * on the synthetic device because modifying the VF device
 771          * statistics will not work correctly.
 772          */
 773         rx_stats = &nvchan->rx_stats;
 774         u64_stats_update_begin(&rx_stats->syncp);
 775         rx_stats->packets++;
 776         rx_stats->bytes += len;
 777
 778         if (skb->pkt_type == PACKET_BROADCAST)
 779                 ++rx_stats->broadcast;
 780         else if (skb->pkt_type == PACKET_MULTICAST)
 781                 ++rx_stats->multicast;
 782         u64_stats_update_end(&rx_stats->syncp);
 783
 784         napi_gro_receive(&nvchan->napi, skb);
 785         return 0;
 786 }
 787
 788 static void netvsc_get_drvinfo(struct net_device *net,
 789                                struct ethtool_drvinfo *info)
 790 {
 791         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 792         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 793 }
 794
 795 static void netvsc_get_channels(struct net_device *net,
 796                                 struct ethtool_channels *channel)
 797 {
 798         struct net_device_context *net_device_ctx = netdev_priv(net);
 799         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 800
 801         if (nvdev) {
 802                 channel->max_combined   = nvdev->max_chn;
 803                 channel->combined_count = nvdev->num_chn;
 804         }
 805 }
 806
 807 static int netvsc_set_channels(struct net_device *net,
 808                                struct ethtool_channels *channels)
 809 {
 810         struct net_device_context *net_device_ctx = netdev_priv(net);
 811         struct hv_device *dev = net_device_ctx->device_ctx;
 812         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 813         unsigned int orig, count = channels->combined_count;
 814         struct netvsc_device_info device_info;
 815         bool was_opened;
 816         int ret = 0;
 817
 818         /* We do not support separate count for rx, tx, or other */
 819         if (count == 0 ||
 820             channels->rx_count || channels->tx_count || channels->other_count)
 821                 return -EINVAL;
 822
 823         if (!nvdev || nvdev->destroy)
 824                 return -ENODEV;
 825
 826         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
 827                 return -EINVAL;
 828
 829         if (count > nvdev->max_chn)
 830                 return -EINVAL;
 831
 832         orig = nvdev->num_chn;
 833         was_opened = rndis_filter_opened(nvdev);
 834         if (was_opened)
 835                 rndis_filter_close(nvdev);
 836
 837         memset(&device_info, 0, sizeof(device_info));
 838         device_info.num_chn = count;
 839         device_info.send_sections = nvdev->send_section_cnt;
 840         device_info.send_section_size = nvdev->send_section_size;
 841         device_info.recv_sections = nvdev->recv_section_cnt;
 842         device_info.recv_section_size = nvdev->recv_section_size;
 843
 844         rndis_filter_device_remove(dev, nvdev);
 845
 846         nvdev = rndis_filter_device_add(dev, &device_info);
 847         if (IS_ERR(nvdev)) {
 848                 ret = PTR_ERR(nvdev);
 849                 device_info.num_chn = orig;
 850                 nvdev = rndis_filter_device_add(dev, &device_info);
 851
 852                 if (IS_ERR(nvdev)) {
 853                         netdev_err(net, "restoring channel setting failed: %ld\n",
 854                                    PTR_ERR(nvdev));
 855                         return ret;
 856                 }
 857         }
 858
 859         if (was_opened)
 860                 rndis_filter_open(nvdev);
 861
 862         /* We may have missed link change notifications */
 863         net_device_ctx->last_reconfig = 0;
 864         schedule_delayed_work(&net_device_ctx->dwork, 0);
 865
 866         return ret;
 867 }
 868
 869 static bool
 870 netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd)
 871 {
 872         struct ethtool_link_ksettings diff1 = *cmd;
 873         struct ethtool_link_ksettings diff2 = {};
 874
 875         diff1.base.speed = 0;
 876         diff1.base.duplex = 0;
 877         /* advertising and cmd are usually set */
 878         ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
 879         diff1.base.cmd = 0;
 880         /* We set port to PORT_OTHER */
 881         diff2.base.port = PORT_OTHER;
 882
 883         return !memcmp(&diff1, &diff2, sizeof(diff1));
 884 }
 885
 886 static void netvsc_init_settings(struct net_device *dev)
 887 {
 888         struct net_device_context *ndc = netdev_priv(dev);
 889
 890         ndc->l4_hash = HV_DEFAULT_L4HASH;
 891
 892         ndc->speed = SPEED_UNKNOWN;
 893         ndc->duplex = DUPLEX_FULL;
 894 }
 895
 896 static int netvsc_get_link_ksettings(struct net_device *dev,
 897                                      struct ethtool_link_ksettings *cmd)
 898 {
 899         struct net_device_context *ndc = netdev_priv(dev);
 900
 901         cmd->base.speed = ndc->speed;
 902         cmd->base.duplex = ndc->duplex;
 903         cmd->base.port = PORT_OTHER;
 904
 905         return 0;
 906 }
 907
 908 static int netvsc_set_link_ksettings(struct net_device *dev,
 909                                      const struct ethtool_link_ksettings *cmd)
 910 {
 911         struct net_device_context *ndc = netdev_priv(dev);
 912         u32 speed;
 913
 914         speed = cmd->base.speed;
 915         if (!ethtool_validate_speed(speed) ||
 916             !ethtool_validate_duplex(cmd->base.duplex) ||
 917             !netvsc_validate_ethtool_ss_cmd(cmd))
 918                 return -EINVAL;
 919
 920         ndc->speed = speed;
 921         ndc->duplex = cmd->base.duplex;
 922
 923         return 0;
 924 }
 925
 926 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 927 {
 928         struct net_device_context *ndevctx = netdev_priv(ndev);
 929         struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
 930         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
 931         struct hv_device *hdev = ndevctx->device_ctx;
 932         int orig_mtu = ndev->mtu;
 933         struct netvsc_device_info device_info;
 934         bool was_opened;
 935         int ret = 0;
 936
 937         if (!nvdev || nvdev->destroy)
 938                 return -ENODEV;
 939
 940         /* Change MTU of underlying VF netdev first. */
 941         if (vf_netdev) {
 942                 ret = dev_set_mtu(vf_netdev, mtu);
 943                 if (ret)
 944                         return ret;
 945         }
 946
 947         netif_device_detach(ndev);
 948         was_opened = rndis_filter_opened(nvdev);
 949         if (was_opened)
 950                 rndis_filter_close(nvdev);
 951
 952         memset(&device_info, 0, sizeof(device_info));
 953         device_info.num_chn = nvdev->num_chn;
 954         device_info.send_sections = nvdev->send_section_cnt;
 955         device_info.send_section_size = nvdev->send_section_size;
 956         device_info.recv_sections = nvdev->recv_section_cnt;
 957         device_info.recv_section_size = nvdev->recv_section_size;
 958
 959         rndis_filter_device_remove(hdev, nvdev);
 960
 961         ndev->mtu = mtu;
 962
 963         nvdev = rndis_filter_device_add(hdev, &device_info);
 964         if (IS_ERR(nvdev)) {
 965                 ret = PTR_ERR(nvdev);
 966
 967                 /* Attempt rollback to original MTU */
 968                 ndev->mtu = orig_mtu;
 969                 nvdev = rndis_filter_device_add(hdev, &device_info);
 970
 971                 if (vf_netdev)
 972                         dev_set_mtu(vf_netdev, orig_mtu);
 973
 974                 if (IS_ERR(nvdev)) {
 975                         netdev_err(ndev, "restoring mtu failed: %ld\n",
 976                                    PTR_ERR(nvdev));
 977                         return ret;
 978                 }
 979         }
 980
 981         if (was_opened)
 982                 rndis_filter_open(nvdev);
 983
 984         netif_device_attach(ndev);
 985
 986         /* We may have missed link change notifications */
 987         schedule_delayed_work(&ndevctx->dwork, 0);
 988
 989         return ret;
 990 }
 991
 992 static void netvsc_get_vf_stats(struct net_device *net,
 993                                 struct netvsc_vf_pcpu_stats *tot)
 994 {
 995         struct net_device_context *ndev_ctx = netdev_priv(net);
 996         int i;
 997
 998         memset(tot, 0, sizeof(*tot));
 999
1000         for_each_possible_cpu(i) {
1001                 const struct netvsc_vf_pcpu_stats *stats
1002                         = per_cpu_ptr(ndev_ctx->vf_stats, i);
1003                 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1004                 unsigned int start;
1005
1006                 do {
1007                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1008                         rx_packets = stats->rx_packets;
1009                         tx_packets = stats->tx_packets;
1010                         rx_bytes = stats->rx_bytes;
1011                         tx_bytes = stats->tx_bytes;
1012                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1013
1014                 tot->rx_packets += rx_packets;
1015                 tot->tx_packets += tx_packets;
1016                 tot->rx_bytes   += rx_bytes;
1017                 tot->tx_bytes   += tx_bytes;
1018                 tot->tx_dropped += stats->tx_dropped;
1019         }
1020 }
1021
1022 static void netvsc_get_stats64(struct net_device *net,
1023                                struct rtnl_link_stats64 *t)
1024 {
1025         struct net_device_context *ndev_ctx = netdev_priv(net);
1026         struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1027         struct netvsc_vf_pcpu_stats vf_tot;
1028         int i;
1029
1030         if (!nvdev)
1031                 return;
1032
1033         netdev_stats_to_stats64(t, &net->stats);
1034
1035         netvsc_get_vf_stats(net, &vf_tot);
1036         t->rx_packets += vf_tot.rx_packets;
1037         t->tx_packets += vf_tot.tx_packets;
1038         t->rx_bytes   += vf_tot.rx_bytes;
1039         t->tx_bytes   += vf_tot.tx_bytes;
1040         t->tx_dropped += vf_tot.tx_dropped;
1041
1042         for (i = 0; i < nvdev->num_chn; i++) {
1043                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1044                 const struct netvsc_stats *stats;
1045                 u64 packets, bytes, multicast;
1046                 unsigned int start;
1047
1048                 stats = &nvchan->tx_stats;
1049                 do {
1050                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1051                         packets = stats->packets;
1052                         bytes = stats->bytes;
1053                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1054
1055                 t->tx_bytes     += bytes;
1056                 t->tx_packets   += packets;
1057
1058                 stats = &nvchan->rx_stats;
1059                 do {
1060                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1061                         packets = stats->packets;
1062                         bytes = stats->bytes;
1063                         multicast = stats->multicast + stats->broadcast;
1064                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1065
1066                 t->rx_bytes     += bytes;
1067                 t->rx_packets   += packets;
1068                 t->multicast    += multicast;
1069         }
1070 }
1071
1072 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1073 {
1074         struct net_device_context *ndc = netdev_priv(ndev);
1075         struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1076         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1077         struct sockaddr *addr = p;
1078         int err;
1079
1080         err = eth_prepare_mac_addr_change(ndev, p);
1081         if (err)
1082                 return err;
1083
1084         if (!nvdev)
1085                 return -ENODEV;
1086
1087         if (vf_netdev) {
1088                 err = dev_set_mac_address(vf_netdev, addr);
1089                 if (err)
1090                         return err;
1091         }
1092
1093         err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1094         if (!err) {
1095                 eth_commit_mac_addr_change(ndev, p);
1096         } else if (vf_netdev) {
1097                 /* rollback change on VF */
1098                 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1099                 dev_set_mac_address(vf_netdev, addr);
1100         }
1101
1102         return err;
1103 }
1104
1105 static const struct {
1106         char name[ETH_GSTRING_LEN];
1107         u16 offset;
1108 } netvsc_stats[] = {
1109         { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1110         { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1111         { "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1112         { "tx_too_big",   offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1113         { "tx_busy",      offsetof(struct netvsc_ethtool_stats, tx_busy) },
1114         { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1115         { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1116         { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1117         { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1118         { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1119 }, vf_stats[] = {
1120         { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1121         { "vf_rx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1122         { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1123         { "vf_tx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1124         { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1125 };
1126
1127 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1128 #define NETVSC_VF_STATS_LEN     ARRAY_SIZE(vf_stats)
1129
1130 /* 4 statistics per queue (rx/tx packets/bytes) */
1131 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4)
1132
1133 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1134 {
1135         struct net_device_context *ndc = netdev_priv(dev);
1136         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1137
1138         if (!nvdev)
1139                 return -ENODEV;
1140
1141         switch (string_set) {
1142         case ETH_SS_STATS:
1143                 return NETVSC_GLOBAL_STATS_LEN
1144                         + NETVSC_VF_STATS_LEN
1145                         + NETVSC_QUEUE_STATS_LEN(nvdev);
1146         default:
1147                 return -EINVAL;
1148         }
1149 }
1150
1151 static void netvsc_get_ethtool_stats(struct net_device *dev,
1152                                      struct ethtool_stats *stats, u64 *data)
1153 {
1154         struct net_device_context *ndc = netdev_priv(dev);
1155         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1156         const void *nds = &ndc->eth_stats;
1157         const struct netvsc_stats *qstats;
1158         struct netvsc_vf_pcpu_stats sum;
1159         unsigned int start;
1160         u64 packets, bytes;
1161         int i, j;
1162
1163         if (!nvdev)
1164                 return;
1165
1166         for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1167                 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1168
1169         netvsc_get_vf_stats(dev, &sum);
1170         for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1171                 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1172
1173         for (j = 0; j < nvdev->num_chn; j++) {
1174                 qstats = &nvdev->chan_table[j].tx_stats;
1175
1176                 do {
1177                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1178                         packets = qstats->packets;
1179                         bytes = qstats->bytes;
1180                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1181                 data[i++] = packets;
1182                 data[i++] = bytes;
1183
1184                 qstats = &nvdev->chan_table[j].rx_stats;
1185                 do {
1186                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1187                         packets = qstats->packets;
1188                         bytes = qstats->bytes;
1189                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1190                 data[i++] = packets;
1191                 data[i++] = bytes;
1192         }
1193 }
1194
1195 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1196 {
1197         struct net_device_context *ndc = netdev_priv(dev);
1198         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1199         u8 *p = data;
1200         int i;
1201
1202         if (!nvdev)
1203                 return;
1204
1205         switch (stringset) {
1206         case ETH_SS_STATS:
1207                 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1208                         memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1209                         p += ETH_GSTRING_LEN;
1210                 }
1211
1212                 for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1213                         memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1214                         p += ETH_GSTRING_LEN;
1215                 }
1216
1217                 for (i = 0; i < nvdev->num_chn; i++) {
1218                         sprintf(p, "tx_queue_%u_packets", i);
1219                         p += ETH_GSTRING_LEN;
1220                         sprintf(p, "tx_queue_%u_bytes", i);
1221                         p += ETH_GSTRING_LEN;
1222                         sprintf(p, "rx_queue_%u_packets", i);
1223                         p += ETH_GSTRING_LEN;
1224                         sprintf(p, "rx_queue_%u_bytes", i);
1225                         p += ETH_GSTRING_LEN;
1226                 }
1227
1228                 break;
1229         }
1230 }
1231
1232 static int
1233 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1234                          struct ethtool_rxnfc *info)
1235 {
1236         const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1237
1238         info->data = RXH_IP_SRC | RXH_IP_DST;
1239
1240         switch (info->flow_type) {
1241         case TCP_V4_FLOW:
1242                 if (ndc->l4_hash & HV_TCP4_L4HASH)
1243                         info->data |= l4_flag;
1244
1245                 break;
1246
1247         case TCP_V6_FLOW:
1248                 if (ndc->l4_hash & HV_TCP6_L4HASH)
1249                         info->data |= l4_flag;
1250
1251                 break;
1252
1253         case UDP_V4_FLOW:
1254                 if (ndc->l4_hash & HV_UDP4_L4HASH)
1255                         info->data |= l4_flag;
1256
1257                 break;
1258
1259         case UDP_V6_FLOW:
1260                 if (ndc->l4_hash & HV_UDP6_L4HASH)
1261                         info->data |= l4_flag;
1262
1263                 break;
1264
1265         case IPV4_FLOW:
1266         case IPV6_FLOW:
1267                 break;
1268         default:
1269                 info->data = 0;
1270                 break;
1271         }
1272
1273         return 0;
1274 }
1275
1276 static int
1277 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1278                  u32 *rules)
1279 {
1280         struct net_device_context *ndc = netdev_priv(dev);
1281         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1282
1283         if (!nvdev)
1284                 return -ENODEV;
1285
1286         switch (info->cmd) {
1287         case ETHTOOL_GRXRINGS:
1288                 info->data = nvdev->num_chn;
1289                 return 0;
1290
1291         case ETHTOOL_GRXFH:
1292                 return netvsc_get_rss_hash_opts(ndc, info);
1293         }
1294         return -EOPNOTSUPP;
1295 }
1296
1297 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1298                                     struct ethtool_rxnfc *info)
1299 {
1300         if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1301                            RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1302                 switch (info->flow_type) {
1303                 case TCP_V4_FLOW:
1304                         ndc->l4_hash |= HV_TCP4_L4HASH;
1305                         break;
1306
1307                 case TCP_V6_FLOW:
1308                         ndc->l4_hash |= HV_TCP6_L4HASH;
1309                         break;
1310
1311                 case UDP_V4_FLOW:
1312                         ndc->l4_hash |= HV_UDP4_L4HASH;
1313                         break;
1314
1315                 case UDP_V6_FLOW:
1316                         ndc->l4_hash |= HV_UDP6_L4HASH;
1317                         break;
1318
1319                 default:
1320                         return -EOPNOTSUPP;
1321                 }
1322
1323                 return 0;
1324         }
1325
1326         if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1327                 switch (info->flow_type) {
1328                 case TCP_V4_FLOW:
1329                         ndc->l4_hash &= ~HV_TCP4_L4HASH;
1330                         break;
1331
1332                 case TCP_V6_FLOW:
1333                         ndc->l4_hash &= ~HV_TCP6_L4HASH;
1334                         break;
1335
1336                 case UDP_V4_FLOW:
1337                         ndc->l4_hash &= ~HV_UDP4_L4HASH;
1338                         break;
1339
1340                 case UDP_V6_FLOW:
1341                         ndc->l4_hash &= ~HV_UDP6_L4HASH;
1342                         break;
1343
1344                 default:
1345                         return -EOPNOTSUPP;
1346                 }
1347
1348                 return 0;
1349         }
1350
1351         return -EOPNOTSUPP;
1352 }
1353
1354 static int
1355 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1356 {
1357         struct net_device_context *ndc = netdev_priv(ndev);
1358
1359         if (info->cmd == ETHTOOL_SRXFH)
1360                 return netvsc_set_rss_hash_opts(ndc, info);
1361
1362         return -EOPNOTSUPP;
1363 }
1364
1365 #ifdef CONFIG_NET_POLL_CONTROLLER
1366 static void netvsc_poll_controller(struct net_device *dev)
1367 {
1368         struct net_device_context *ndc = netdev_priv(dev);
1369         struct netvsc_device *ndev;
1370         int i;
1371
1372         rcu_read_lock();
1373         ndev = rcu_dereference(ndc->nvdev);
1374         if (ndev) {
1375                 for (i = 0; i < ndev->num_chn; i++) {
1376                         struct netvsc_channel *nvchan = &ndev->chan_table[i];
1377
1378                         napi_schedule(&nvchan->napi);
1379                 }
1380         }
1381         rcu_read_unlock();
1382 }
1383 #endif
1384
1385 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1386 {
1387         return NETVSC_HASH_KEYLEN;
1388 }
1389
1390 static u32 netvsc_rss_indir_size(struct net_device *dev)
1391 {
1392         return ITAB_NUM;
1393 }
1394
1395 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1396                            u8 *hfunc)
1397 {
1398         struct net_device_context *ndc = netdev_priv(dev);
1399         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1400         struct rndis_device *rndis_dev;
1401         int i;
1402
1403         if (!ndev)
1404                 return -ENODEV;
1405
1406         if (hfunc)
1407                 *hfunc = ETH_RSS_HASH_TOP;      /* Toeplitz */
1408
1409         rndis_dev = ndev->extension;
1410         if (indir) {
1411                 for (i = 0; i < ITAB_NUM; i++)
1412                         indir[i] = rndis_dev->rx_table[i];
1413         }
1414
1415         if (key)
1416                 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1417
1418         return 0;
1419 }
1420
1421 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1422                            const u8 *key, const u8 hfunc)
1423 {
1424         struct net_device_context *ndc = netdev_priv(dev);
1425         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1426         struct rndis_device *rndis_dev;
1427         int i;
1428
1429         if (!ndev)
1430                 return -ENODEV;
1431
1432         if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1433                 return -EOPNOTSUPP;
1434
1435         rndis_dev = ndev->extension;
1436         if (indir) {
1437                 for (i = 0; i < ITAB_NUM; i++)
1438                         if (indir[i] >= ndev->num_chn)
1439                                 return -EINVAL;
1440
1441                 for (i = 0; i < ITAB_NUM; i++)
1442                         rndis_dev->rx_table[i] = indir[i];
1443         }
1444
1445         if (!key) {
1446                 if (!indir)
1447                         return 0;
1448
1449                 key = rndis_dev->rss_key;
1450         }
1451
1452         return rndis_filter_set_rss_param(rndis_dev, key);
1453 }
1454
1455 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1456  * It does have pre-allocated receive area which is divided into sections.
1457  */
1458 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1459                                    struct ethtool_ringparam *ring)
1460 {
1461         u32 max_buf_size;
1462
1463         ring->rx_pending = nvdev->recv_section_cnt;
1464         ring->tx_pending = nvdev->send_section_cnt;
1465
1466         if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1467                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1468         else
1469                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1470
1471         ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1472         ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1473                 / nvdev->send_section_size;
1474 }
1475
1476 static void netvsc_get_ringparam(struct net_device *ndev,
1477                                  struct ethtool_ringparam *ring)
1478 {
1479         struct net_device_context *ndevctx = netdev_priv(ndev);
1480         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1481
1482         if (!nvdev)
1483                 return;
1484
1485         __netvsc_get_ringparam(nvdev, ring);
1486 }
1487
1488 static int netvsc_set_ringparam(struct net_device *ndev,
1489                                 struct ethtool_ringparam *ring)
1490 {
1491         struct net_device_context *ndevctx = netdev_priv(ndev);
1492         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1493         struct hv_device *hdev = ndevctx->device_ctx;
1494         struct netvsc_device_info device_info;
1495         struct ethtool_ringparam orig;
1496         u32 new_tx, new_rx;
1497         bool was_opened;
1498         int ret = 0;
1499
1500         if (!nvdev || nvdev->destroy)
1501                 return -ENODEV;
1502
1503         memset(&orig, 0, sizeof(orig));
1504         __netvsc_get_ringparam(nvdev, &orig);
1505
1506         new_tx = clamp_t(u32, ring->tx_pending,
1507                          NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1508         new_rx = clamp_t(u32, ring->rx_pending,
1509                          NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1510
1511         if (new_tx == orig.tx_pending &&
1512             new_rx == orig.rx_pending)
1513                 return 0;        /* no change */
1514
1515         memset(&device_info, 0, sizeof(device_info));
1516         device_info.num_chn = nvdev->num_chn;
1517         device_info.send_sections = new_tx;
1518         device_info.send_section_size = nvdev->send_section_size;
1519         device_info.recv_sections = new_rx;
1520         device_info.recv_section_size = nvdev->recv_section_size;
1521
1522         netif_device_detach(ndev);
1523         was_opened = rndis_filter_opened(nvdev);
1524         if (was_opened)
1525                 rndis_filter_close(nvdev);
1526
1527         rndis_filter_device_remove(hdev, nvdev);
1528
1529         nvdev = rndis_filter_device_add(hdev, &device_info);
1530         if (IS_ERR(nvdev)) {
1531                 ret = PTR_ERR(nvdev);
1532
1533                 device_info.send_sections = orig.tx_pending;
1534                 device_info.recv_sections = orig.rx_pending;
1535                 nvdev = rndis_filter_device_add(hdev, &device_info);
1536                 if (IS_ERR(nvdev)) {
1537                         netdev_err(ndev, "restoring ringparam failed: %ld\n",
1538                                    PTR_ERR(nvdev));
1539                         return ret;
1540                 }
1541         }
1542
1543         if (was_opened)
1544                 rndis_filter_open(nvdev);
1545         netif_device_attach(ndev);
1546
1547         /* We may have missed link change notifications */
1548         ndevctx->last_reconfig = 0;
1549         schedule_delayed_work(&ndevctx->dwork, 0);
1550
1551         return ret;
1552 }
1553
1554 static const struct ethtool_ops ethtool_ops = {
1555         .get_drvinfo    = netvsc_get_drvinfo,
1556         .get_link       = ethtool_op_get_link,
1557         .get_ethtool_stats = netvsc_get_ethtool_stats,
1558         .get_sset_count = netvsc_get_sset_count,
1559         .get_strings    = netvsc_get_strings,
1560         .get_channels   = netvsc_get_channels,
1561         .set_channels   = netvsc_set_channels,
1562         .get_ts_info    = ethtool_op_get_ts_info,
1563         .get_rxnfc      = netvsc_get_rxnfc,
1564         .set_rxnfc      = netvsc_set_rxnfc,
1565         .get_rxfh_key_size = netvsc_get_rxfh_key_size,
1566         .get_rxfh_indir_size = netvsc_rss_indir_size,
1567         .get_rxfh       = netvsc_get_rxfh,
1568         .set_rxfh       = netvsc_set_rxfh,
1569         .get_link_ksettings = netvsc_get_link_ksettings,
1570         .set_link_ksettings = netvsc_set_link_ksettings,
1571         .get_ringparam  = netvsc_get_ringparam,
1572         .set_ringparam  = netvsc_set_ringparam,
1573 };
1574
1575 static const struct net_device_ops device_ops = {
1576         .ndo_open =                     netvsc_open,
1577         .ndo_stop =                     netvsc_close,
1578         .ndo_start_xmit =               netvsc_start_xmit,
1579         .ndo_set_rx_mode =              netvsc_set_multicast_list,
1580         .ndo_change_mtu =               netvsc_change_mtu,
1581         .ndo_validate_addr =            eth_validate_addr,
1582         .ndo_set_mac_address =          netvsc_set_mac_addr,
1583         .ndo_select_queue =             netvsc_select_queue,
1584         .ndo_get_stats64 =              netvsc_get_stats64,
1585 #ifdef CONFIG_NET_POLL_CONTROLLER
1586         .ndo_poll_controller =          netvsc_poll_controller,
1587 #endif
1588 };
1589
1590 /*
1591  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1592  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1593  * present send GARP packet to network peers with netif_notify_peers().
1594  */
1595 static void netvsc_link_change(struct work_struct *w)
1596 {
1597         struct net_device_context *ndev_ctx =
1598                 container_of(w, struct net_device_context, dwork.work);
1599         struct hv_device *device_obj = ndev_ctx->device_ctx;
1600         struct net_device *net = hv_get_drvdata(device_obj);
1601         struct netvsc_device *net_device;
1602         struct rndis_device *rdev;
1603         struct netvsc_reconfig *event = NULL;
1604         bool notify = false, reschedule = false;
1605         unsigned long flags, next_reconfig, delay;
1606
1607         /* if changes are happening, comeback later */
1608         if (!rtnl_trylock()) {
1609                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1610                 return;
1611         }
1612
1613         net_device = rtnl_dereference(ndev_ctx->nvdev);
1614         if (!net_device)
1615                 goto out_unlock;
1616
1617         rdev = net_device->extension;
1618
1619         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1620         if (time_is_after_jiffies(next_reconfig)) {
1621                 /* link_watch only sends one notification with current state
1622                  * per second, avoid doing reconfig more frequently. Handle
1623                  * wrap around.
1624                  */
1625                 delay = next_reconfig - jiffies;
1626                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1627                 schedule_delayed_work(&ndev_ctx->dwork, delay);
1628                 goto out_unlock;
1629         }
1630         ndev_ctx->last_reconfig = jiffies;
1631
1632         spin_lock_irqsave(&ndev_ctx->lock, flags);
1633         if (!list_empty(&ndev_ctx->reconfig_events)) {
1634                 event = list_first_entry(&ndev_ctx->reconfig_events,
1635                                          struct netvsc_reconfig, list);
1636                 list_del(&event->list);
1637                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1638         }
1639         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1640
1641         if (!event)
1642                 goto out_unlock;
1643
1644         switch (event->event) {
1645                 /* Only the following events are possible due to the check in
1646                  * netvsc_linkstatus_callback()
1647                  */
1648         case RNDIS_STATUS_MEDIA_CONNECT:
1649                 if (rdev->link_state) {
1650                         rdev->link_state = false;
1651                         netif_carrier_on(net);
1652                         netif_tx_wake_all_queues(net);
1653                 } else {
1654                         notify = true;
1655                 }
1656                 kfree(event);
1657                 break;
1658         case RNDIS_STATUS_MEDIA_DISCONNECT:
1659                 if (!rdev->link_state) {
1660                         rdev->link_state = true;
1661                         netif_carrier_off(net);
1662                         netif_tx_stop_all_queues(net);
1663                 }
1664                 kfree(event);
1665                 break;
1666         case RNDIS_STATUS_NETWORK_CHANGE:
1667                 /* Only makes sense if carrier is present */
1668                 if (!rdev->link_state) {
1669                         rdev->link_state = true;
1670                         netif_carrier_off(net);
1671                         netif_tx_stop_all_queues(net);
1672                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
1673                         spin_lock_irqsave(&ndev_ctx->lock, flags);
1674                         list_add(&event->list, &ndev_ctx->reconfig_events);
1675                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1676                         reschedule = true;
1677                 }
1678                 break;
1679         }
1680
1681         rtnl_unlock();
1682
1683         if (notify)
1684                 netdev_notify_peers(net);
1685
1686         /* link_watch only sends one notification with current state per
1687          * second, handle next reconfig event in 2 seconds.
1688          */
1689         if (reschedule)
1690                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1691
1692         return;
1693
1694 out_unlock:
1695         rtnl_unlock();
1696 }
1697
1698 static struct net_device *get_netvsc_bymac(const u8 *mac)
1699 {
1700         struct net_device *dev;
1701
1702         ASSERT_RTNL();
1703
1704         for_each_netdev(&init_net, dev) {
1705                 if (dev->netdev_ops != &device_ops)
1706                         continue;       /* not a netvsc device */
1707
1708                 if (ether_addr_equal(mac, dev->perm_addr))
1709                         return dev;
1710         }
1711
1712         return NULL;
1713 }
1714
1715 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1716 {
1717         struct net_device *dev;
1718
1719         ASSERT_RTNL();
1720
1721         for_each_netdev(&init_net, dev) {
1722                 struct net_device_context *net_device_ctx;
1723
1724                 if (dev->netdev_ops != &device_ops)
1725                         continue;       /* not a netvsc device */
1726
1727                 net_device_ctx = netdev_priv(dev);
1728                 if (!rtnl_dereference(net_device_ctx->nvdev))
1729                         continue;       /* device is removed */
1730
1731                 if (rtnl_dereference(net_device_ctx->vf_netdev) == vf_netdev)
1732                         return dev;     /* a match */
1733         }
1734
1735         return NULL;
1736 }
1737
1738 /* Called when VF is injecting data into network stack.
1739  * Change the associated network device from VF to netvsc.
1740  * note: already called with rcu_read_lock
1741  */
1742 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
1743 {
1744         struct sk_buff *skb = *pskb;
1745         struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
1746         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1747         struct netvsc_vf_pcpu_stats *pcpu_stats
1748                  = this_cpu_ptr(ndev_ctx->vf_stats);
1749
1750         skb->dev = ndev;
1751
1752         u64_stats_update_begin(&pcpu_stats->syncp);
1753         pcpu_stats->rx_packets++;
1754         pcpu_stats->rx_bytes += skb->len;
1755         u64_stats_update_end(&pcpu_stats->syncp);
1756
1757         return RX_HANDLER_ANOTHER;
1758 }
1759
1760 static int netvsc_vf_join(struct net_device *vf_netdev,
1761                           struct net_device *ndev)
1762 {
1763         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1764         int ret;
1765
1766         ret = netdev_rx_handler_register(vf_netdev,
1767                                          netvsc_vf_handle_frame, ndev);
1768         if (ret != 0) {
1769                 netdev_err(vf_netdev,
1770                            "can not register netvsc VF receive handler (err = %d)\n",
1771                            ret);
1772                 goto rx_handler_failed;
1773         }
1774
1775         ret = netdev_upper_dev_link(vf_netdev, ndev, NULL);
1776         if (ret != 0) {
1777                 netdev_err(vf_netdev,
1778                            "can not set master device %s (err = %d)\n",
1779                            ndev->name, ret);
1780                 goto upper_link_failed;
1781         }
1782
1783         /* set slave flag before open to prevent IPv6 addrconf */
1784         vf_netdev->flags |= IFF_SLAVE;
1785
1786         schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
1787
1788         call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
1789
1790         netdev_info(vf_netdev, "joined to %s\n", ndev->name);
1791         return 0;
1792
1793 upper_link_failed:
1794         netdev_rx_handler_unregister(vf_netdev);
1795 rx_handler_failed:
1796         return ret;
1797 }
1798
1799 static void __netvsc_vf_setup(struct net_device *ndev,
1800                               struct net_device *vf_netdev)
1801 {
1802         int ret;
1803
1804         /* Align MTU of VF with master */
1805         ret = dev_set_mtu(vf_netdev, ndev->mtu);
1806         if (ret)
1807                 netdev_warn(vf_netdev,
1808                             "unable to change mtu to %u\n", ndev->mtu);
1809
1810         if (netif_running(ndev)) {
1811                 ret = dev_open(vf_netdev);
1812                 if (ret)
1813                         netdev_warn(vf_netdev,
1814                                     "unable to open: %d\n", ret);
1815         }
1816 }
1817
1818 /* Setup VF as slave of the synthetic device.
1819  * Runs in workqueue to avoid recursion in netlink callbacks.
1820  */
1821 static void netvsc_vf_setup(struct work_struct *w)
1822 {
1823         struct net_device_context *ndev_ctx
1824                 = container_of(w, struct net_device_context, vf_takeover.work);
1825         struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
1826         struct net_device *vf_netdev;
1827
1828         if (!rtnl_trylock()) {
1829                 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
1830                 return;
1831         }
1832
1833         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
1834         if (vf_netdev)
1835                 __netvsc_vf_setup(ndev, vf_netdev);
1836
1837         rtnl_unlock();
1838 }
1839
1840 static int netvsc_register_vf(struct net_device *vf_netdev)
1841 {
1842         struct net_device *ndev;
1843         struct net_device_context *net_device_ctx;
1844         struct netvsc_device *netvsc_dev;
1845
1846         if (vf_netdev->addr_len != ETH_ALEN)
1847                 return NOTIFY_DONE;
1848
1849         /*
1850          * We will use the MAC address to locate the synthetic interface to
1851          * associate with the VF interface. If we don't find a matching
1852          * synthetic interface, move on.
1853          */
1854         ndev = get_netvsc_bymac(vf_netdev->perm_addr);
1855         if (!ndev)
1856                 return NOTIFY_DONE;
1857
1858         net_device_ctx = netdev_priv(ndev);
1859         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1860         if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
1861                 return NOTIFY_DONE;
1862
1863         if (netvsc_vf_join(vf_netdev, ndev) != 0)
1864                 return NOTIFY_DONE;
1865
1866         netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
1867
1868         dev_hold(vf_netdev);
1869         rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
1870         return NOTIFY_OK;
1871 }
1872
1873 /* VF up/down change detected, schedule to change data path */
1874 static int netvsc_vf_changed(struct net_device *vf_netdev)
1875 {
1876         struct net_device_context *net_device_ctx;
1877         struct netvsc_device *netvsc_dev;
1878         struct net_device *ndev;
1879         bool vf_is_up = netif_running(vf_netdev);
1880
1881         ndev = get_netvsc_byref(vf_netdev);
1882         if (!ndev)
1883                 return NOTIFY_DONE;
1884
1885         net_device_ctx = netdev_priv(ndev);
1886         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1887         if (!netvsc_dev)
1888                 return NOTIFY_DONE;
1889
1890         netvsc_switch_datapath(ndev, vf_is_up);
1891         netdev_info(ndev, "Data path switched %s VF: %s\n",
1892                     vf_is_up ? "to" : "from", vf_netdev->name);
1893
1894         return NOTIFY_OK;
1895 }
1896
1897 static int netvsc_unregister_vf(struct net_device *vf_netdev)
1898 {
1899         struct net_device *ndev;
1900         struct net_device_context *net_device_ctx;
1901
1902         ndev = get_netvsc_byref(vf_netdev);
1903         if (!ndev)
1904                 return NOTIFY_DONE;
1905
1906         net_device_ctx = netdev_priv(ndev);
1907         cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
1908
1909         netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
1910
1911         netdev_rx_handler_unregister(vf_netdev);
1912         netdev_upper_dev_unlink(vf_netdev, ndev);
1913         RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
1914         dev_put(vf_netdev);
1915
1916         return NOTIFY_OK;
1917 }
1918
1919 static int netvsc_probe(struct hv_device *dev,
1920                         const struct hv_vmbus_device_id *dev_id)
1921 {
1922         struct net_device *net = NULL;
1923         struct net_device_context *net_device_ctx;
1924         struct netvsc_device_info device_info;
1925         struct netvsc_device *nvdev;
1926         int ret = -ENOMEM;
1927
1928         net = alloc_etherdev_mq(sizeof(struct net_device_context),
1929                                 VRSS_CHANNEL_MAX);
1930         if (!net)
1931                 goto no_net;
1932
1933         netif_carrier_off(net);
1934
1935         netvsc_init_settings(net);
1936
1937         net_device_ctx = netdev_priv(net);
1938         net_device_ctx->device_ctx = dev;
1939         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1940         if (netif_msg_probe(net_device_ctx))
1941                 netdev_dbg(net, "netvsc msg_enable: %d\n",
1942                            net_device_ctx->msg_enable);
1943
1944         hv_set_drvdata(dev, net);
1945
1946         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1947
1948         spin_lock_init(&net_device_ctx->lock);
1949         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
1950         INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
1951
1952         net_device_ctx->vf_stats
1953                 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
1954         if (!net_device_ctx->vf_stats)
1955                 goto no_stats;
1956
1957         net->netdev_ops = &device_ops;
1958         net->ethtool_ops = &ethtool_ops;
1959         SET_NETDEV_DEV(net, &dev->device);
1960
1961         /* We always need headroom for rndis header */
1962         net->needed_headroom = RNDIS_AND_PPI_SIZE;
1963
1964         /* Initialize the number of queues to be 1, we may change it if more
1965          * channels are offered later.
1966          */
1967         netif_set_real_num_tx_queues(net, 1);
1968         netif_set_real_num_rx_queues(net, 1);
1969
1970         /* Notify the netvsc driver of the new device */
1971         memset(&device_info, 0, sizeof(device_info));
1972         device_info.num_chn = VRSS_CHANNEL_DEFAULT;
1973         device_info.send_sections = NETVSC_DEFAULT_TX;
1974         device_info.send_section_size = NETVSC_SEND_SECTION_SIZE;
1975         device_info.recv_sections = NETVSC_DEFAULT_RX;
1976         device_info.recv_section_size = NETVSC_RECV_SECTION_SIZE;
1977
1978         nvdev = rndis_filter_device_add(dev, &device_info);
1979         if (IS_ERR(nvdev)) {
1980                 ret = PTR_ERR(nvdev);
1981                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1982                 goto rndis_failed;
1983         }
1984
1985         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1986
1987         /* hw_features computed in rndis_netdev_set_hwcaps() */
1988         net->features = net->hw_features |
1989                 NETIF_F_HIGHDMA | NETIF_F_SG |
1990                 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
1991         net->vlan_features = net->features;
1992
1993         netdev_lockdep_set_classes(net);
1994
1995         /* MTU range: 68 - 1500 or 65521 */
1996         net->min_mtu = NETVSC_MTU_MIN;
1997         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
1998                 net->max_mtu = NETVSC_MTU - ETH_HLEN;
1999         else
2000                 net->max_mtu = ETH_DATA_LEN;
2001
2002         ret = register_netdev(net);
2003         if (ret != 0) {
2004                 pr_err("Unable to register netdev.\n");
2005                 goto register_failed;
2006         }
2007
2008         return ret;
2009
2010 register_failed:
2011         rndis_filter_device_remove(dev, nvdev);
2012 rndis_failed:
2013         free_percpu(net_device_ctx->vf_stats);
2014 no_stats:
2015         hv_set_drvdata(dev, NULL);
2016         free_netdev(net);
2017 no_net:
2018         return ret;
2019 }
2020
2021 static int netvsc_remove(struct hv_device *dev)
2022 {
2023         struct net_device_context *ndev_ctx;
2024         struct net_device *vf_netdev;
2025         struct net_device *net;
2026
2027         net = hv_get_drvdata(dev);
2028         if (net == NULL) {
2029                 dev_err(&dev->device, "No net device to remove\n");
2030                 return 0;
2031         }
2032
2033         ndev_ctx = netdev_priv(net);
2034
2035         netif_device_detach(net);
2036
2037         cancel_delayed_work_sync(&ndev_ctx->dwork);
2038
2039         /*
2040          * Call to the vsc driver to let it know that the device is being
2041          * removed. Also blocks mtu and channel changes.
2042          */
2043         rtnl_lock();
2044         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2045         if (vf_netdev)
2046                 netvsc_unregister_vf(vf_netdev);
2047
2048         unregister_netdevice(net);
2049
2050         rndis_filter_device_remove(dev,
2051                                    rtnl_dereference(ndev_ctx->nvdev));
2052         rtnl_unlock();
2053
2054         hv_set_drvdata(dev, NULL);
2055
2056         free_percpu(ndev_ctx->vf_stats);
2057         free_netdev(net);
2058         return 0;
2059 }
2060
2061 static const struct hv_vmbus_device_id id_table[] = {
2062         /* Network guid */
2063         { HV_NIC_GUID, },
2064         { },
2065 };
2066
2067 MODULE_DEVICE_TABLE(vmbus, id_table);
2068
2069 /* The one and only one */
2070 static struct  hv_driver netvsc_drv = {
2071         .name = KBUILD_MODNAME,
2072         .id_table = id_table,
2073         .probe = netvsc_probe,
2074         .remove = netvsc_remove,
2075 };
2076
2077 /*
2078  * On Hyper-V, every VF interface is matched with a corresponding
2079  * synthetic interface. The synthetic interface is presented first
2080  * to the guest. When the corresponding VF instance is registered,
2081  * we will take care of switching the data path.
2082  */
2083 static int netvsc_netdev_event(struct notifier_block *this,
2084                                unsigned long event, void *ptr)
2085 {
2086         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2087
2088         /* Skip our own events */
2089         if (event_dev->netdev_ops == &device_ops)
2090                 return NOTIFY_DONE;
2091
2092         /* Avoid non-Ethernet type devices */
2093         if (event_dev->type != ARPHRD_ETHER)
2094                 return NOTIFY_DONE;
2095
2096         /* Avoid Vlan dev with same MAC registering as VF */
2097         if (is_vlan_dev(event_dev))
2098                 return NOTIFY_DONE;
2099
2100         /* Avoid Bonding master dev with same MAC registering as VF */
2101         if ((event_dev->priv_flags & IFF_BONDING) &&
2102             (event_dev->flags & IFF_MASTER))
2103                 return NOTIFY_DONE;
2104
2105         switch (event) {
2106         case NETDEV_REGISTER:
2107                 return netvsc_register_vf(event_dev);
2108         case NETDEV_UNREGISTER:
2109                 return netvsc_unregister_vf(event_dev);
2110         case NETDEV_UP:
2111         case NETDEV_DOWN:
2112                 return netvsc_vf_changed(event_dev);
2113         default:
2114                 return NOTIFY_DONE;
2115         }
2116 }
2117
2118 static struct notifier_block netvsc_netdev_notifier = {
2119         .notifier_call = netvsc_netdev_event,
2120 };
2121
2122 static void __exit netvsc_drv_exit(void)
2123 {
2124         unregister_netdevice_notifier(&netvsc_netdev_notifier);
2125         vmbus_driver_unregister(&netvsc_drv);
2126 }
2127
2128 static int __init netvsc_drv_init(void)
2129 {
2130         int ret;
2131
2132         if (ring_size < RING_SIZE_MIN) {
2133                 ring_size = RING_SIZE_MIN;
2134                 pr_info("Increased ring_size to %u (min allowed)\n",
2135                         ring_size);
2136         }
2137         netvsc_ring_bytes = ring_size * PAGE_SIZE;
2138         netvsc_ring_reciprocal = reciprocal_value(netvsc_ring_bytes);
2139
2140         ret = vmbus_driver_register(&netvsc_drv);
2141         if (ret)
2142                 return ret;
2143
2144         register_netdevice_notifier(&netvsc_netdev_notifier);
2145         return 0;
2146 }
2147
2148 MODULE_LICENSE("GPL");
2149 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2150
2151 module_init(netvsc_drv_init);
2152 module_exit(netvsc_drv_exit);